Some impressions of our open lab day on the 20th of May 2022:

Some impressions of our hiking day:

Student Assistant at the Montanuniversität Leoben

Short bio:Tanja Sukal, B.Sc. started at CPS in October  2024.

Tanja Sukal is a master student at Montanuniversität Leoben in the program Industrial Data Science. Her research interests include data modeling based on machine learning algorithms including neural networks. 

Research Interests

• Data Modeling 
• Machine Learning
• Python Programming

Past Thesis

• B.Sc. Thesis: Sukal Tanja on Creating a Python development environment for LEGO Ev3 robot systems, Defended on April 26, 2024

Contact

Tanja Sukal
Student Assistant at the Chair of Cyber-Physical-Systems
Montanuniversität Leoben
Franz-Josef-Straße 18,
8700 Leoben, Austria 

Email: tanja.sukal@stud.unileoben.ac.at

Meeting 06/07

Research

• Investigating Representation Collapse in Reinforcement Learning Agents from Vision
  • plan/structure?
  • what RL algorithms?
    • visual data
    • Gehart Neumann, Marc Toussaint, Joustus Piater (Innsburg)
    • Define a research question
    • Focus on some domain
    •  
• Unnormalized Contrastive learning
  • All CL models use l2 normalization of the representation

    • Stability: Normalizing the representations ensures that they all have the same magnitude. This can make the learning process more stable, as it prevents the model from assigning arbitrarily large or small magnitudes to the representations.

    • Focus on direction: By constraining the representations to have a fixed magnitude, the learning process focuses on the direction of the vectors in the embedding space. This is often what we care about in tasks like contrastive learning, where the goal is to make the representations of similar inputs point in similar directions.

    • Computational convenience: As mentioned earlier, many computations, such as the dot product between two vectors, are easier to perform and interpret in normalized spaces.

    • Interpretability: Normalized representations are often more interpretable, as the angle between two vectors can be directly interpreted as a measure of similarity or dissimilarity.

  • BUT, this come to the expense of
    • Decreased Capacity: With normalization, the model’s capacity to represent data is reduced since it can only rely on the direction of vectors in the embedding space. This limitation may result in the model being less able to capture complex patterns in the data.

• • • Missing Magnitude Information: The absence of magnitude information in normalized vectors removes the ability to convey meaningful data properties such as confidence levels or other relevant characteristics. Normalization discards this information, limiting the model’s understanding of the data.
  • IDEA: remove the l2 regularization
    • Regularize the model to penalize large magnitudes.
    • Scale the representations to a desired range.
    • Design a custom loss function considering both direction and magnitude
• Breaking Binary: Towards a Continuum of Conceptual Similarities in Self-Supervised Learning
  • will take more time to set-up
  • will leave it for later

PhD Registration

• registered

M.Sc. Students/Interns

• Iye Szin presenting next week her work until now.

ML Course

• Publish Video Tutorial on pytorch

Miscellaneous

• Summer School in Cambridge
  • Poster?

Meeting 25/07

Research

• Goal-oriented working mode:
  • define subgoals and milestones
  • (make sure that you can evaluate them, and define criteria of success, scores, etc.)
  • till 17.08.2023 10:00
• Define topic, sub-problem, open challenge, your approach, toy task, full experiment
• RAAD2024, 20.12.2023 concept paper with first results
• Spring 2024 A+ robotics conference paper on simulation experiments.
• Summer 2024 A+ robotics paper on real robot experiments

M.Sc. Students/Interns

Miscellaneous

Laufende Projekte, Bachelor- und Masterarbeiten

• Retrofitting of a Cyber-Physical System to a reactive molding machine for thermoset resins

Meeting 15/06

Research

• reviews for ECAI (2/6) (Vedant is working on one of them)
• Research leads:
  1. Dimensionality Collapse of Visual Representations in Reinforcement Learning
  2. Improve SwAV architecture by creating better latent space clusters with the use of Sparse Autoencoders

PhD Registration

• waiting for admission office response

M.Sc. Students/Interns

• Iye Szin has a working prototype

ML Course

• Tutorial on pytorch
• pending grading for assignments 5 and 6

Miscellaneous

• Summer School
  • Registration done
  • Air tickets booked
  • accommodation booked
• English course got postponed

Meeting 22/06

Research

• Reviews for ECAI 2023 done.

M.Sc. Students/Interns

ML Course

Miscellaneous

Meeting 29/06

Research

M.Sc. Students/Interns

ML Course

Miscellaneous

Meeting on the 19th of July 2023

Agenda

Location: Scholz Rohstoffhandel, Industriestraße 11, 2361 Laxenburg

Date & Time: 19th June 2023, 11am to 12pm

Participants: Melanie Neubauer, M.Sc.

Top 1: Notes

• Scholz processes about 30 thousand tons of scrap, about 1 million tons in Austria per year. Sources of scrap: consumers, industry, collection yards of municipalities, repair shops.
• Aluminium, Messing, Kupfer, Kunststoffe (Gummi), Draht – are the problematic components. Bateries and accumulators are also in the scrap and need to be sorted out. Copper is the most important part that must be removed, as this greatly affects the quality of the steel.

• Scrap consists of end-of-life vehicles and whole or crushed and from old household appliances. The cars are disassembled beforehand (tires, fluids, engine, etc.)
• This scrap is used to produce the so-called E40 material, for which there are EU guidelines, but the requirements of the steel producers are higher. The demands of steel and plastics manufacturers must be met. (single-variety) (EU – 2.5cm, Scholz – 10-15cm)
• Criteria in the steel plant: radioactivity, weight, visibility (copper!, plastics, rubber, substances)
• For the E40 material, 1 car and then 2 portions of household scrap are always put into the shredder. The 1:2 blends look about the same as just car scrap. Mixing is done to prevent canting in the shredder. By means of an eddy current separator (use air), the shredder output is divided into the light fraction and the E40 metal fraction. With the help of manual re-sorting, the quality of both outputs is improved. The re-sorting takes place on the two conveyor belts (light fraction, E40 fraction).
• The cameras for data acquisition should be implemented in the area at the hand sorting. If this does not work, a bypass can be created. Also tests can be carried out in the technical center with approx. 5000 stk scrap (Alexia says this is a suitable quantity). 
• technical center in St. Michael owns: NIR, WIS, laser triangulation 
• From next spring, cars will be more partially dismantled before shredding, there will be a re-data collection.

• Libs sensor: very complex and lengthy, has already been carried out at Scholz, silicon and magnesium are detected by it, laser shoots particles out of scrap particles and analyzes the composition
• MBT material: That waste which is removed from the household residual waste with the help of a magnet. This waste goes directly into the shredder, then metal is sorted out again, the rest goes back to the incinerator. (currently not relevant)

Meeting on the 20th of July 2023

Agenda

Location: Chair of CPS

Date & Time: 20th July 2023, 8am to 9am

Participants: Univ.-Prof. Dr. Elmar Rueckert, Melanie Neubauer, M.Sc.

1. Organisatorial progress update by Melanie.
2. Next steps of Melanie.

Top 1: Organisational Update

Update to the Visit to Scholz from 19th of July. (Images are in the Cloud under Projects/Kiramet)

Top 2: Next steps

• Paper for RAAD conference to be submitted. Deadline for submission is 12/20/2023. 
• This should focus on one/ or more methods for segmentation of problematic particles in scrap. 
• Topic: application of classification and segmentation to scrap particles. 
• Methods used: Mask R-CNN, …
• The classes that occur are to be described. 
• In the chapter Analyses will be analyzed how much of what is present in the sample in percent.
• First of all the paper should be written down in note form.
• Chapter: Introduction, Methods, Analyses, Summary….

Master Thesis Student at the Montanuniversität Leoben

Short bio: Klemens is an Energy Engineering student at Montanuniversität Leoben,  working on a Master’s Thesis named “Deep Neural Energy Forecasting for  
Economic Resource Usage in Hydrogen Industries”. This work focuses on  exploring how AI can be used to better manage resources in the hydrogen industry.

Klemens got his start in Electrical Engineering, graduating from a  technical secondary school. After a brief but interesting stint with the Military Orchestra in Carinthia, he decided to return to his  engineering roots, earning a Bachelor of Science in Raw Materials Engineering.

Now, as a Master’s candidate, Klemens hopes to combine his skills and  interests to make a positive contribution to the energy sector.

Research Interests

• Deep Learning
• Resource utilization in Energy Sector

Thesis

•  
• Supervision: Vedant Dave, M.Sc.

Contact

Klemens Lechner, B. Sc
Master Thesis Student at the Chair of Cyber-Physical-Systems
Montanuniversität Leoben
Franz-Josef-Straße 18,
8700 Leoben, Austria 

Email: klemens.lechner@stud.unileoben.ac.at

Meeting on the 5th of June 2023

Agenda

Location: Chair of CPS

Date & Time: 5th June 2023, 11am to 12pm

Participants: Univ.-Prof. Dr. Elmar Rueckert, Melanie Neubauer, M.Sc.

1. Organisatorial progress update by Melanie.
2. Infos from Elmar regarding the progress.
3. Next steps of Melanie.

Top 1: Update Melanie

•  Moodle access works
•  Visit St. Michael with my own car on 13.06.
• Home Office on 14.06.
• Problems with the SparseDetr Code on my Laptop (running the code is not possible), running the code in the lab on the 3090 is possible

Top 2: Infos from Elmar

• In St michael extra garbage is delivered to record data
• Alexia has examples of images of the garbage
• Not much open source data for the project in St. Michael available
• Possibility to run the DMS paper Code, and use it for Linus’ Roboter

Top 3: Next Steps

•  Check references from the project paper (E-Mail Elmar)
• Train SparseDetr on the 3090 GPU (check how it works)
• DMS paper Apple (https://machinelearning.apple.com/research/dense-material)
  try to get access to the data and run the code
• Read Paper “Applications of convolutional neural networks for intelligent waste identification on recicling: A review” (interesting information about datasets,…) 

Meeting on the 13th of June 2023

Agenda

Location: Chair of CPS

Date & Time: 13th June 2023, 11am to 12pm

Participants: Univ.-Prof. Dr. Elmar Rueckert, Melanie Neubauer, M.Sc.

1. Organisatorial progress update by Melanie.
2. Feedback to the research talk presentation by Elmar.
3. Topics of promising future research direction.
4. Next steps of Melanie.
5. Date of the next meeting.

Top 1: Organisational Update

add you text here. 

Meeting on the 23th of May 2023

Location: Chair of CPS

Date & Time: 23th May 2023, 10:30 am to 11:00 am

Participants: Univ.-Prof. Dr. Elmar Rueckert, Melanie Neubauer, M.Sc.

Agenda

1. Organisatorial progress update by Melanie.
2. Topics of promising future research direction.
3. Next steps of Melanie.

Discussion regarding Recycling Lab St. Michael

• wait for E-Mail from Alexia Tischberger-Aldrian (visit in St. Michael)
• currently no data available
• server is set up to collect data
• various sensors record the data
• Infos regarding data – see Project Manual p. 16 chapter 7.1

Discussion regarding future publications

1. Publication
– publicate the collected and labeled data from St. Michael
– generate a GUI for labeling the data 
– the labeling is made by study assistants (about 200.000 Images)

2. Publication
– Conference Paper about Particle Tracking
– Train a network on the basis of the first Publication

eventually 3. Publication Transfer Learning
– use Open source Data for training

Main Question: How does a network learn an efficient representation to be able to build a reasonable model (even with a small amount of training data)?

To Do:

• wait for E-Mail from Alexia Tischberger-Aldrian to visit the Recycling Lab
• Ask for Waste-RL Dataset from ‘Waste Management’ Paper (www.elsevier.com/locate/wastman) – Images from Waste-RL are not very suitable
• maybe find some other Dataset instead of Waste-RL
•  Read Paper Waste Management
• Use Paper ‘Sparse Detr…’ on the Steel Defect Dataset from the Master Thesis